Notes from BIO-Europe 2026 in Cologne

I’m back from BIO-Europe, and I learned something very important.
In Cologne, the waiter keeps bringing Kölsch until you put a beer mat over your glass.
A visibly empty glass triggers an automatic replacement. And, like loops, this can continue indefinitely (or until you fall asleep on the floor, I suppose).
AI in pharma works much the same way, except that many large companies have reached for the beer mat before the first round.
That turned out to be the theme of the week.
Two things struck me in the partnering meetings and the corridors. First, AI was in every conversation, in the decks, and increasingly in the meetings itself, taking notes.
Second, in a surprising number of those conversations, some people from large companies told me they still can’t use general purpose AI tools on real work.
So AI is everywhere. And AI is nowhere. Both statements are true, and the space between them is where most of us now work.
What follows are some examples of AI that I heard about while in Cologne, along with some additional examples of where our industry is and is not using AI.
AI Is Everywhere
The drug discovery story has been told to death, so let’s look elsewhere.
AI has quietly moved into the parts of pharma that run on documents, meetings and decisions; essentially everywhere.
The boardroom. In 2023, Sanofi announced it was going “all in” on artificial intelligence, with the stated ambition of becoming the first pharma company powered by AI at scale. Three years on, Sanofi says its agentic app, Plai, aggregates more than a billion data points and participates in all governance meetings where molecule progression decisions are made.
The regulatory dossier. Novo Nordisk built NovoScribe, a documentation platform for clinical study reports and related documents, using Claude Code on Amazon Bedrock. Anthropic’s case study claims that time spent producing clinical study documentation fell from more than ten weeks to ten minutes.
That’s the vendor’s number, so take that claim for what it’s worth.
Meanwhile, other accounts cite more modest (but still very large) gains. Either way, in September the two companies announced a broader R&D collaboration, with Novo’s CEO describing an ambition to become the world’s most AI-driven healthcare company.
The marketing department. Pfizer built “Charlie”, a generative AI platform for its content supply chain that also helps with fact-checking and legal review. Anyone who has waited six months for CDA sign-off will understand the appeal.
Everyone else. Merck has expanded its internal generative AI platform to thousands of employees, including tools for clinical study report writing, and Bayer’s internal platform reportedly serves around 40,000 employees.
The regulator. Even the FDA is in on it. The agency reported that generative AI use among its own staff grew from about 1% in early 2025 to over 80% by May 2026. So even if you don’t use AI to write your submission, there’s a fair chance an AI will help read it.
And in BD, of course, there are the notetakers.
At BIO-Europe, we spotted everything from laptop and mobile notetakers to one wristwatch-style recorder.
At no point did anyone ask if the meeting could be recorded.
More on those later.
AI Is Nowhere
Confidentiality. The best-known data point is a 2024 ZoomRx survey in which more than half of life sciences professionals said their companies had banned ChatGPT, including 65% of the top 20 pharmas. Nearly all cited security and the potential to leak internal data.
That survey is two and a half years old, which is a geological era in AI time. But judging from what I heard in Cologne, the instinct hasn’t gone away. It has simply become more polite.
Notably, the same survey found that more than half of respondents were using ChatGPT in some form anyway.
A ban, it turns out, is not the same thing as non-use.
Sovereignty. For European companies, there is an additional layer. The US CLOUD Act allows American authorities to compel US-based technology companies to hand over data, regardless of where it is stored. Meanwhile, a March 2026 Politico survey found that 84% of Europeans don’t trust American tech companies to handle their personal data responsibly.
To be fair, the evidence also suggests that most European businesses still prioritize AI adoption and productivity over full sovereignty.
Concern and adoption can coexist, but they can lead to awkward budgeting and procurement meetings.
Cost. AI runs on electricity, and Europe’s electricity is expensive. According to IEA figures cited by White & Case, prices for energy-intensive industries in Europe were, on average, about double those in the US and 50% higher than in China and India.
Compute is not free anywhere, but it is less free in Frankfurt, Germany versus Frankfurt, Kentucky.
And now the EU is taxing the internet? I heard about this on my way to Cologne, so I looked into it.
The short answer is no, but there are two real debates behind the rumour, and they are worth understanding.
The first is the Digital Networks Act. European telecom operators have long argued that the large platforms generating most internet traffic should help pay for network infrastructure, the so-called “fair share” or network fee debate. In January 2026, the Commission declined to force Big Tech to pay for the telecoms overhaul. Instead, it proposed a voluntary conciliation mechanism, which some tech companies fear could eventually lead to mandatory network fees.
That fear is not purely theoretical.
The Parliament’s lead MEP has since proposed amendments under which Amazon, Google and Meta could face binding regulatory decisions if interconnection talks with operators fail, and the Internet Society has urged MEPs to reject them.
Complicating matters, the 2025 US–EU trade agreement includes an EU pledge not to introduce network fees. A committee vote is planned for the first quarter of 2027.
Telecom operators and Big Tech have now been arguing about who pays for the pipes for longer than most Phase 3 trials take to read out, and with no end in sight.
The second is the EU budget. In April 2026, the European Parliament adopted a position on the 2028–2034 budget that calls for a digital services levy on major digital platforms, among other new revenue sources.
The Commission had left a bloc-wide digital services tax out of its own proposal, but has since estimated that a digital levy could raise about €5 billion a year.
Neither of these is a tax on pharma’s internet traffic, and neither has been adopted. If either survives, it would be reasonable to expect some of the cost to flow downstream into cloud and AI pricing. But for a European CFO already looking at high power costs and sovereignty questions, it is one more reason to wait and see.
AI in the Grey Zone
Between “everywhere” and “nowhere” lies a grey zone, and that is where BD&L professionals actually live and work. Three examples:
The uninvited guest. AI notetakers are standard in virtual meetings, and they are creeping into in-person ones.
They are also attracting lawyers.
In August 2025, a proposed class action was filed against Otter.ai, alleging that its notetaker recorded conversations without the consent of all participants and used them to train its models.
Now apply that to a partnering meeting.
You are pitching a non-confidential version of your asset, with no CDA yet in place.
The other side’s bot is listening.
Where does the transcript go? Whose servers? Is it training anything?
Is it summarizing your answer to the off-the-cuff question about your CMC issues without adequate data and context?
What if you say something that is not entirely correct? How does that impact how the bot interprets the conversation?
Interesting times we’re living in.
The lopsided table. A small biotech can use whatever AI tools it likes. With younger companies, they have grown up using Claude or Codex or even Notebook LLM. However, its big-pharma counterpart may be restricted to Copilot, an internal platform, or to nothing at all.
What This Means
I’ve written before about what AI means for BD professionals, so I’ll keep this short.
- Assume a machine reads your deck first. Make the non-confidential deck say only what you would be comfortable seeing in someone else’s database. Then make it easy for both machine and human to summarize. Slides with six graphs and pie charts will be misinterpreted by the LLM, and that is potentially costly.
- Ask about the notetaker. It’s not rude. A simple “Is anything recording this meeting, and where does it go?” is now a reasonable question, especially before a CDA is signed. Be ready to answer it about your own tools, too.
- Know your counterpart’s AI policy. It tells you how fast they can move, who will actually read your materials, and what they can and cannot do with them. If a CDA is contemplated, this should be discussed prior to execution.
- Don’t over-read the headlines. The EU is not taxing your internet traffic. It is having a long, loud argument about who pays for networks and how to fund its budget. Watch it, but don’t build your 2027 plan around it.
- Keep a human accountable. Whatever the tool, someone still has to own the answer. Regulators, Boards and Judges remain stubbornly uninterested in “the AI said so.”
AI is everywhere, and AI is nowhere. Most deals, as usual, will be done somewhere in between.
In the meantime, try the Kölsch, and this Bullsh!t wine.
The glasses are small, but they keep coming.
Which is roughly how AI adoption in pharma is going.